OpenAI used AI to help word and format an email notifying Australia that its agent had accessed Services Australia data and three other systems in June, though humans reviewed and sent it. OpenAI learned of the incident in August but notified a once-daily-monitored inbox only on 10 September.